I’ve been quiet for the past month. In part because I was busy, and in part because I was tired of reporting on states manipulating their case counts by suppressing testing.
In the last few weeks, judging from the national news, there’s been a spike in cases nationally despite this gamesmanship. Frankly, it was kind of inevitable, though I wasn’t sure it would be obvious prior to the election.
It’s an easily knowable number, so this morning I spent a couple of hours calculating it.
All calculations are based on the Covid Tracking Project’s data. For this purpose I compared the latest 2-week period (10/24 to 10/10) to the prior 2-week period (10/09 to 09/25). My original scope for this analysis was the 50 states plus DC. However, there were some unusual adjustments that distorted the results for Hawaii and Missouri during this period, so I eliminated them.
Here’s what happened: Latest 2 weeks v Prior 2 weeks for 48 states + DC (excluding HI and MO)….
Total Tests: 12.9 million v 11.9
Positive tests (new cases): 817,309 v 597,073
Positivity (positive tests/(total tests-pending)): 6.3% v 5.0%
Period over period increase: 220,236 Increase due to Positivity increase: 165,770 Increase due to Testing increase: 54,466
Overall there are 17 states with a positivity rate >10%. Of these, only one state (Wisconsin) showed a decrease in positivity between the two periods. The state with the highest positivity in the current period is ND, at an astonishing 41.9%.
20 states experienced positivity between 5.0% and 10%. Only two (DE & OR) experienced a decrease between the two periods.
The Twitter thread I referenced above has 5 clips from the Meadows interview. It’s clear that the administration has completely surrendered any pretense of trying to control the pandemic, short of a vaccine.
Florida’s peak week for the number of tests and new cases ended on July 17. Even though Positivity on that week was a staggering 19.0%, which screams-out for additional testing, the state cut back on test counts the next week, and for the next five consecutive weeks. Between July 17 and August 28, weekly testing levels dropped by 60%.
It wasn’t until the week ending September 5, in the face of increased positivity, that Florida’s testing counts increased, but modestly, still 53% below the July 17 peak.
Trump has been widely ridiculed in the press for saying that “more testing leads to more cases”. In the common sense meaning of these words that may seem like a stupid statement, but add one word and it becomes completely true: “more testing leads to more reported cases”.
In this post we’re going to estimate the impact of Florida’s decision to cut testing counts in reducing the Confirmed Case count. Here we calculated an alternative scenario where testing remained at the July 17 levels. As you can see in the chart, Floridas actual testing peaked at 459,151, and dropped steadily to 180,465 for the week ending 28 August, and then increased slightly to 216,130.
If Florida had maintained peak testing levels, how many additional cases does this add? That depends on the Positivity Level, i.e. the percent of people tested who test positive. We calculated using the CDC recommended methodology (see this post for a deep dive on how Florida plays games with the Positivity Level it reports on its dashboard).
To estimate additional Cases, we need to estimate the positivity level on the incremental tests. Generally speaking, when the number of tests goes up, you expect the positivity level to go down somewhat, on the theory that the sickest people tend to get tested first, so that additional tests hit a slightly healthier population.
In Florida’s case, this hasn’t always been true. Florida allocates tests not by need, but by county population. This means that tests don’t flow preferentially to the counties with the highest infection rate. Indeed, in a recent county level analysis, we discovered that the counties with the lowest Positivity Rate were actually testing more than the counties with the highest rates. (see Table 2 in this post). Were Florida to increase testing and change its allocation policies so that counties with higher positivity rates received more tests, there’s a reasonable chance Florida’s average Positivity rate would actually increase even as testing counts increased.
Nevertheless, we think it’s unlikely that Florida would change its allocation methods to one that would increase reported case counts. As a result, we decided to assume a similar allocation based on county population alone, and estimated the Positivity as as shown in Figure 2.
The Scenario Estimate in Figure 2 is calculated as follow:
The original tests are carried forward at the observed Positivity rate
Incremental tests within the scenario are estimated at a lower Positivity rate that decreases 10% for each 10% increment over the number of actual tests. In other words, the first 10% of incremental tests (as a % of actual tests for the week) are carried forward at 90% of the observed Positivity rate, the next 10% at 80%, and so forth.
As shown in Figure 3, for the week ending 8/28, the state actually tested only 180,645 people (with a reported Positivity Rate of 12.5%). In our scenario, we “tested” an additional 276,626 and calculated an incremental Positivity rate of 8.0% using our formula. This converted to a total of 459,151 tests at an average positivity rate of 9.8%.
Overall, in the 7 weeks since the peak week of July 17, the State of Florida reported 318,626 cases, instead of the 463,454 we estimated would have been generated had testing levels remained constant. This means that Florida under-reported new cases by more than 46%.
Based on this analysis, there remains some improvement in Florida’s situation. But it’s not nearly as large as normally reported. Note that the lower case count will ripple into other parts of the reporting. In particular, deaths resulting from cases that are not confirmed will NOT be reported in Florida’s official COVID counts, as Florida (contrary to CDC recommendations) does not report probable COVID deaths.
Part 2 – Manipulating Positivity and Reported Deaths
The Trump campaign spews misinformation, trying to convince the electorate that the administration’s response to Corona-19 has been effective. In Part 1, we focused on how (mostly) Republican governors are limiting testing in order to reduce “Confirmed Cases” (Please read first, if you haven’t already).
In that post, we introduced this equation:
(a) [Confirmed Cases] = [People Tested] x [% positive]
Using Florida as a case study, we focused on how the state has reduced its Confirmed Case Count by cutting back dramatically on the number of [People Tested], at a time when best-practice suggests increased testing is warranted.
In this post, we’ll focus on the second variable, [% positive], aka “positivity”, and how it is being manipulated to obscure the reporting of the pandemic, including official reporting of death counts. As we’ll see, these manipulations are creating the appearance of progress against the Pandemic in the run up to the election, but masking the true impact of the disease, and potentially risking thousands of additional deaths.
Why Test at All? A Scenario
There are many reasons for testing, but the most common (and important) is to identify people who are infectious, to isolate themselves, and thereby breaking the cycle of contagion.
Consider this scenario:
Your best friend calls you to say that he’s been diagnosed with COVID. Six days earlier, you hosted him for dinner at your home. Your friend started to feel sick the day after the dinner, procrastinated a day thinking it was allergies, then got tested. Three days later, today, he gets a positive result and he’s called you immediately to give you the bad news.
You’d been reasonably responsible during the dinner party, which was mostly outdoors (though it was a still evening with no wind). He’d spent half an hour conversing with you in the kitchen as you prepared the meats for the BBQ, not always masked, everyone had used the same downstairs powder room, and while you ate outdoors, you drank wine, and talked until the small hours, only a few feet apart.
Should you get tested? I certainly would want to, even if I felt no symptoms, and I suspect most folks would agree, particularly if you’re in a high-risk category (as I am due to age).
So, you go to the testing center and 48 hours later get the good news, negative. However, a few days later you start coughing. So, you get tested again and the result this time is positive.
How does this get reported? According to CDC guidelines, the two tests would both be recorded on the case reporting form, but in the testing summary reports, you’d be listed as a single person tested, with a single positive result. Your personal “positivity” would be 100%.
As you recover, you get additional tests to help determine if you’re still communicable. After 10 days you’re still testing positive, but that result would not change your positivity score. Nor, after another week, when you test negative, does that undo your earlier positive score. It all gets grouped within your case file (and a reason why a Confirmed Case is a thing).
It makes sense.
Most states report positivity following these guidelines. But, Florida mostly does not.
You can figure out the conventional positivity number if you dig into the DOH County Report and calculate it yourself. You can also download the Covid Tracking Project’s reporting, which aggregates the same numbers to the state level. This allows you to compare Florida to other states.
Calculated the CDC-recommended way, Florida’s positivity was 13.9% for the week ending 8/23. This is an important number for several reasons, but specifically, positivity above 10% is considered a dangerous level, and a sign that more testing is needed.
However, if you go onto the DOH’s COVID-19 Data and Surveillance Dashboard, it will tell you that the positivity for the same week ending 8/23 was 5.83%. Huh?
Turns out that Florida does its own, non-standard, and widely-criticized positivity calculation. Essentially, the state counts only a single positive result but, contrary to CDC guidelines, counts all of the individual negative results. These controversial measures also appear on the County Report as the so called “New” Test/Pos/Neg. If you calculate the way the State of Florida does on its dashboard, for the scenario we just discussed, your personal positivity would be 33% not 100%:
The initial negative result
The one positive result (but not the second)
The final negative result
I can see no reason to do this other than to confuse the public, and allow Florida officials to claim a lower-sounding positivity rate. Most folks won’t understand the difference, so that if the state says “positivity” is less than 6%, and many other Sun Belt states are struggling with 10%+, Florida must be doing better, right? The fact that the Trump administration won’t enforce CDC reporting guidelines is another example of how the Administration is placing politics above the national interest.
We saw in Part 1 that Florida was cutting back on total testing numbers in order to reduce the number of confirmed cases. Given the continuing high positivity number (as the CDC would calculate it), you can be sure that there are many people who would like to be tested who aren’t. The question is, are the scarce tests being allocated fairly among the people who need them? Spoiler answer, no.
Allocating Testing in a Time of Scarcity
So how should tests be allocated? Absent scarcity or pricing barriers (i.e. if testing were free), you’d expect test demand to correlate strongly with the number of sick people.
So, an optimal scheme for allocating tests to counties would take into account both the base population of the county, and the positivity rate, in order to allocate more tests to the places where they’re needed.
Florida, instead, appears to allocate testing capacity solely by population. This has the advantage of being bureaucratically defensible, and fairly straight-forward, but leads to some less than optimal results.
As shown in Table 1, for the week ending July 17, counties with above-average positivity tested only 5% more than counties with below-average positivity. For the week of August 27, it was even worse, where the counties with higher than average positivity only managed 3% more testing.
Table 2 breaks this down for counties with the highest and lowest Positivity rates for 8/27. Column 5 of this table provides a Positivity Ratio of the 8/27 and 7/17 percentages. A ratio >1.0 means that positivity increased, <1.0, that it dropped. As a state-wide average, Florida’s Positivity dropped to 12.45% on August 27 from 17.89% on July 17, a 0.7 ratio.
Despite the drop state-wide, 18 of the 20 highest-ratio counties on 8/27 had experienced a Positivity increase. By contrast, 10 out of 10 of the lowest Positivity experienced a decrease at the state-average-rate or better.
In terms of testing rates, counties with Positivity Ranks 11-20 were particularly badly served by the state’s allocation methods, achieving only 589 tests/100k during the week. Equally unfair, the 10 counties at the bottom of the Positivity Ranking were able to test the most at 871/100k, 19.4% higher than the state-average testing level of 729/100k, and 47.9% higher than Counties 11-20 could achieve. This is far from optimal.
Why does the state allocate tests so inflexibly? I’m speculating. But realize that if the state were to allocate extra tests to high-positivity counties it would translate to a higher Confirmed Case count.
Impact on CFR
The Trump administration puts a big emphasis on the Case Fatality Rate, i.e. the percentage of Confirmed Cases who eventually died from COVID.
We did a deep dive on CFR a couple of weeks ago (Is FL 5x better than NJ?) which I don’t want to repeat here but is well worth reading. However, it’s worth noting a few key points:
By reducing Confirmed Cases, the State of Florida is also reducing Confirmed Deaths
Florida doesn’t follow CDC guidelines and does NOT report Probable COVID deaths as part of its official total. So, this misinformation is lowering the official death count substantially.
This means that many Florida COVID deaths won’t be reported explicitly, and will only be understood over time, as excess deaths are identified and analyzed. This is slow process, especially if death reporting is trickling in. It will always be subject to gas-lighting and denial by the political powers that be.
In mid-August, I estimated probable COVID Deaths for Florida as an additional 47% on top of Confirmed Deaths. This was based on the CDC’s excess death estimates in July and early August, based mostly on Confirmed Cases created during a time when Florida was NOT purposefully cutting back on testing. Based on this current testing and reporting practices, excess deaths are likely to increase markedly. Keep in mind, however, it won’t be clear until mid-October at the earliest, given the long lag time between infection and death, particularly for the younger patients that Florida is infecting.
The State of Florida needs only keep-up this fiction, that the situation is improving, viable through the election. I hope this article will encourage more folks to dig deeper and expose what’s really going on.
The pandemic in mid-July, 2020, most agree, was the worst of times in the Sun Belt. But by late August, the Trump campaign is trying desperately to convince the American public and media that, if not now quite the best of times, things are improving.
Some states are supporting this narrative with deceptive reporting of critical COVID statistics. Unfortunately, it’s working, providing a misleading picture of progress against this disease while endangering lives in the process.
As an example of a misleadingly optimistic report, see this extended analysis published by Will Feuer on CNBC News website on August 25. A brief excerpt:
“Fauci’s worst fears have not come to pass as daily new cases have steadily fallen across much of the U.S….
In Florida, the daily average number of new cases has fallen from about 11,100 [per day] on July 22 to about 3,900 [per day] this week. Cindy Prins, an epidemiologist at the University of Florida, attributed much of the drop to changing behavior across the state, prompted by news coverage and effective public health messaging.”
CNBC News, August 25, 2020
Feuer is at least somewhat aware that “Confirmed Cases” can be manipulated. But, along with most of the MSM, he doesn’t do the work to tease apart the official truth-bending, even in a heavily reported piece. As we’ll see, according to the numbers, changing behavior in Florida actually had very little to do with the reduction in Confirmed Cases.
To be clear, I am not an epidemiologist. But I am a former management consultant, who spent a fair number of years working with highly dysfunctional large corporations that in some cases institutionalized misleading reporting (Note 1). The skills I developed then are useful here in ferreting out what’s going on with states like Florida, which is, in my opinion, publishing politically-motivated, official misinformation about the status of the pandemic.
What is a Confirmed Case?
Trump was widely ridiculed for saying the “more testing leads to more cases”. In point of fact, it’s an accurate statement if you add one word: “more testing leads to more reported cases”. This subtle difference seems lost on many reporters covering this story. It’s not lost on Red governors looking to support the President’s campaign narrative.
In the situation the US finds itself today, a “Confirmed Case” is an artificial construct. Recognize that a Confirmed Case means something far different from the common-sense meaning of the words, which would be someone sick with an “infection”. The number of infections, estimated by the CDC and others, is likely 10x higher than the current confirmed case count. So, of course, more testing leads to more cases.
When states are acting in good faith to attack the problem, and reporting consistently, Confirmed Cases are an OK barometer for the status of the pandemic. But, because (as we’ll see) the numbers are so easy to manipulate, it’s highly problematic if the state is acting in bad faith.
To illustrate how this works, I’m going to do a deep dive on Florida in two parts. Part 1 will look at testing numbers. Part 2 (which I’ll publish later this week) will take a close look at positivity (the percentage of tests confirming an infection) and how this is being manipulated.
The Undisputed Facts
In mid-July, Florida was reeling from the Pandemic. It had assumed the mantle as COVID’s epicenter in the US from New York. The press narrative was all about how out-of-control things were, and particularly focused on DeSantis’ mishandling of the pandemic.
By the end of August, the narrative is switching to Florida getting things under control.
Here are the “top-line” numbers. The week ending July 17 is arguably Florida’s worst week; while the week ending August 27 appears to show substantial, continuing improvement:
Yes, it’s true that New Cases fell by 74% between these two periods, from 10,000/day to less than 2,700 . Most reporting goes no deeper than this. What’s wrong with that?
A Plan to Hide the Truth?
What’s wrong is it points to a story that isn’t true. Florida is showing slight signs of improvement, perhaps, but the pandemic is far from under control. However, the state’s top line reporting paints a far rosier picture. To get to the truth, you need to dig deeper and calculate some numbers yourself (something few reporters seem to do).
In Florida’s case, you’ll want to download the daily Florida DOH County Report (Note 2). This is the source document for the official case count, and therefore deserves close examination. The document was originally designed so that, the number of positive tests = confirmed cases, based on the number of people tested. Put another way, you can think of the report as driving a formula calculating Confirmed Cases as follows:
(a) [Confirmed Cases] = [People Tested] x [% positive] Note: “% positive” is often referred to as “Positivity”, a term we’ll adopt for the rest of this post. (a) is simply a label.
You need to dig this deep because the State of Florida plays games with its top-line positivity reporting (and publishes alternative testing counts that we’ll discuss in detail in Part 2).
For example, for the week ending July 19, the DOH Dashboard was reporting positivity as 12.73%. However, based on the people-based testing counts provided on the County Report, the positivity was 17.89% for the week ending July 17. We’ve adopted these County Report variables because they are the source for the underlying DOH case count. They were also designed consistent with the methodology for case and test reporting that was recommended by the pre-politicized CDC, and which can be compared to most other states (Note 3).
By this calculation, the County Report tells you that Total People Tested fell from 406,435 during the Week ending 7/17, to 150,259 for the week of 8/27, a 63% drop!
Feuer, in his article, is aware that Case Count is influenced by testing counts. As he states:
While testing has declined in recent weeks, the number of new cases is falling faster than testing rates, indicating that at least some of the drop is real….
When you dive into the numbers, new cases fell by 54,025 for the weeks ending July 17 to August 27. But, based on the formula (a), Cases would have fallen by 45,849 simply due to reduced testing numbers, even if Positivity had stayed flat. By this reckoning, 85% of the reduction is due directly to reduced test counts!
Total Tests (per 100k)
Feuer doesn’t run the Florida numbers himself, and seems to accept Cindy Prins’ explanation (quoted at top) at face value, that “much of the drop” was due to changing behavior. However, behavior impacts positivity, explaining only 15% of the drop. The other 85% is due to the State of Florida’s testing levels. I note that Feuer quotes Dr. Prins, an Epidemiologist at the University of Florida, indirectly, so the quote may not be accurate. She is, however, an employee of the State of Florida.
Is there a benign explanation for cutting testing?
A positivity rate of 17.89% is dangerously high. To be experiencing this rate and be cutting back on testing suggests either gross incompetence or callous disregard for human life (or both).
For example, see this graph which contrasts New York’s testing history to Florida’s. Note these are both large states (population nearly the same); both have been hit hard by the Pandemic. New York was hit earlier, but has been the most successful state in terms of reducing new infections and caseloads. Florida was hit later, and has done a much worse job in controlling the infection.
Once Florida’s case rate started to climb in mid-June, Florida appeared to follow the New York model, increasing testing dramatically until July 17. However, New York has continued to increase its testing rate even as positivity has fallen to under 1%.
By contrast, Florida began cutting the number of tests week by week starting July 18, so that by the week ending August 27th it was testing at less than half of its peak level, falling to to 44th out of 51 states (including DC).
Even the rate of 1,973/100k that Florida was testing at the mid-July peak isn’t particularly high. It would have ranked 8th during the week of August 27th. The rate of 729 seems shockingly low: CA @ 5.64% tested at 1,489.
Most of the states you see with high positivity (>10%) and low testing rates (<1,000/100k) are other states presumably supporting the “Pandemic is Over” narrative of the Trump campaign. Texas, struggling with a similar positivity rate to Florida’s is at 14.1% positivity and is testing at 745/100k. Others, starting with the lowest testing levels, are SC @ 28.44%/292; ID @ 12.88%/735; MO @ 14.09%/766; SD @ 17.45%/851; MS @ 16.29%/899; IA @ 16.77%/903; and KS @ 10.70%/989. The only “blue” state meeting the same criteria (positivity> 10%, testing < 1,000/100k) is NV @ 15.55%/592.
We also took a close look at Florida’s testing rate by county. We’ll have a LOT more to say about this in Part II of this post, but we think it’s worth introducing here. The maps show the level of testing from the County Report, expressed in Tests-per-Person / per 100,000 population for the weeks ending 7/17 and 8/27, respectively.
On average across the entire state, the testing level fell by 63%. However, the two counties with the highest rate in the state are:
Leon County, where the State Capital is located, testing at 1,980/100k. The testing rate actually rose during this interval, the only county where this happened, from 1,535 during the week of July 17.
Alachua County, home of the University of Florida, at a rate of 1,562/100k, a reduction of less than 20% period-to-period.
It would seem the state government, while willing to endanger its citizens in support of a narrative, is taking care of itself!
Note 1: From 1986 to 1993, I was a consultant with the New York office of McKinsey & Company. My first engagement was with the computer division of a major company that by the accounting of the McKinsey team was losing $2 billion/year, and by the company’s books, about $300 million. Playing games with numbers is not restricted to states.
Note 2: The sources for most of this analysis are the Cases by County dataset published by the Florida Department of Health. This allowed me to calculate case and test reporting at a granular level for the weeks ending 7/17 and 8/27. Not precisely the dates quoted by Feuer, but close enough. This data is archived on the University of Florida’s Florida COVID-19 Hub, which you can find here. For the purposes of weekly totals, I downloaded tables for 7/11, 7/17, 8/21, and 8/27, to allow me to calculate the difference in cumulative totals.
Note 3: The CDC’s original, non-political guidelines recommended reporting by people tested rather than samples, recognizing that an individual can receive multiple tests prior to testing positive. When health practitioners talk about Positivity, the percentages they quote usually assume tests reported this way. Reporting based on samples will always show a lower percentage than reporting based on people. We’ll explain this in more detail in Part II (Florida’s alternative approach actually mixes the two methods, making it even more confusing).
President Trump loves to cite the statistic Case Fatality Rate (“CFR”) as proof the US response to the pandemic is the “best in the world”. His choice is not accidental. If you look carefully, he’s cherry-picked just about the only statistic where the US doesn’t look awful relative to most other advanced countries. He suggests this difference is because US hospitals are more effective than other countries. It’s a lie.
CFR is actually a terrible “top-line” measure for pandemic response effectiveness. The ultimate goal is to keep people from dying or becoming disabled. If you can keep people from getting sick in the first place, it’s much better than curing them, as they don’t suffer, don’t risk permanent health impairment or death, and society avoids spending resources for treatments that are no longer necessary.
If you want a single statistic to measure response-effectiveness (as Jonathan Swan pointed out in his famous HBO/Axios interview), Deaths/100,000 population is much better than CFR. But on this measure the US looks mediocre in comparison to many other countries. And you still have to take into account likely future deaths (not just those which have already occurred). Recognizing the infections are spreading in the US faster than any other advanced country, the US’ death rate is likely to end up among the worst. So, of course, Trump ignores it.
But, even setting all of that aside, the US “lead” in CFR is largely illusory. To illustrate why, we’ll contrast Florida’s CFR to New Jersey’s, a state whose battle with the pandemic resembles the pattern in many of the early-hit European countries, much more than Florida.
What is CFR?
CFR is a simple ratio: [Deaths] / [Cases]. A “naive” CFR calculation looks like this:
Source: COVID Tracking Project 08/12/2020
Calculated this way, it turns out that New Jersey has the second highest CFR in the US (after Connecticut). No wonder Trump loves to talk about Florida when he promotes his views on CFR.
Do you believe that Florida’s hospitals are 5x better than New Jersey’s in curing COVID? Or is something else going on here?
The short answer
The short answer is, “Yes! A lot of things are going on.” We’ll expand on each of these points, but here are the key explanations for the differences:
Florida cases are made up of significantly younger people than New Jersey’s. This translates directly to a lower death rate.
Florida’s cases are newer than New Jersey’s, so a higher percentage have yet to die.
Florida is currently generating new cases at a much higher rate, temporarily lowering CFR even more.
New Jersey reports both confirmed and probable COVID deaths, while Florida reports only confirmed deaths. This significantly lowers the apparent CFR for Florida.
New Jersey was unlucky. Or, you could say, badly located, or made serious mistakes (really, all of the above). Given the once-in-100 year nature of this pandemic, imo, they amount to nearly the same thing.
Florida Cases are Younger
Younger people die from COVID at a much lower rate than older people. In New Jersey, 30-year-old’s died at a rate of 0.22%. Folks aged 80+ have died at a rate of 37%! (Note 1)
A major reason for Florida’s lower CFR is its median age for Confirmed Cases is about 10 years younger than New Jersey’s. As shown in the table, if we apply the New Jersey CFR for each age segment to Florida’s distribution, New Jersey’s overall CFR would decline from 7.7% (as it stood on July 30) to 4.8%.
% NJ Cases
% FL Cases
NJ Age Distribution of Cases and Deaths per NJ Department of Health updated 7/30/20 Florida Case Distributions from Florida Covid Action Master File 8/14/2020. FL age categories converted to NJ categories using a straight-line annual distribution within each FL age-segment.
Note that many European countries have significantly older populations than the US. It’s clear that comparing aggregate CFR from one jurisdiction to another without understanding the underlying age distribution can be very misleading.
More Florida cases haven’t yet died
As of 8/12, the average New Jersey case was 104 days old, while the average Florida case was 37 days.
Once a case shows up in the statistics, it takes a surprisingly long time for the corresponding deaths to occur and be reported. Half of case fatalities take place within 15 days from the date they’re first recorded. Add 7 days delay for the death to show up in the reports, and you’re up to 22 days for half of the deaths to be reported. The next 25% take another week, but then the final 25% trickle in over about another month. For our analysis, we assumed 100% of deaths would be counted after 57 days. (Note 2)
Based on this distribution, of the 557,000 Confirmed Florida Cases on 8/12, 201,000 (36.1%) had yet to resolve into either a death or a recovery. Contrast that to NJ where only only 9,213 (4.9%) are unresolved. Source: pandemic-sense.com original analysis based on Covid Tracking Project Data from 8/12/20.
Most European countries are in a similar situation to New Jersey, with a much older set of cases, which are mostly resolved, in comparison to the US.
Florida is currently generating new cases at a high rate
While related to the previous point, a high rate of new cases also inflates the denominator of the ratio, reducing the CFR even more, albeit only temporarily. Florida added over 48,000 cases in the 7 days from 8/6-8/12: 8.6% of its total cases to date. By contrast, New Jersey added only 2,611 during the same period, 1.4% of its total.
Unlike deaths, which take a while, Florida’s 48,000 cases hit the denominator immediately,reducing apparent CFR. As long as Florida continues to add cases at a high rate, its CFR will always appear lower than its true rate.
New Jersey is experiencing the opposite phenomenon. On July 31, CFR was 7.7%. Over the next two weeks, deaths from old cases continued to trickle in, while relatively very few new cases were added. As a result, CFR increased to 8.5%.
Does that mean NJ is doing a worse job treating its COVID patients? Of course not. Anyone successfully dealing with the virus is eventually going to experience an end-phase when CFR will increase. It’s another reason why it’s stupid to rely on CFR as a sole measure of success.
Florida Doesn’t Report Probable Deaths
The CDC recommends that states report both probable and confirmed deaths in their COVID statistics. New Jersey does so, Florida does not.
What’s the difference? A “Confirmed Death” means a positive COVID test result was received. A “Probable Death” means the deceased was diagnosed by a medical professional as dying from COVID, but not confirmed by a test.
New Jersey added 1,854 probable deaths to its tally on June 25, and continues to report probables, now no differently from other deaths.
Florida does NOT follow the CDC guidelines and reports only confirmed deaths. Realize that this is part of a consistent pattern by Florida’s government to obfuscate what’s actually going on in the state. It gives them the flexibility to cut back on testing (thereby managing the apparent case growth) without CFR exploding. See our post, Faked Out, Not Fake News
To try to estimate the impact, we took a careful look at the CDC’s published “Excess Death” rates, shown in the chart below.
In the chart, each green bar represents total deaths from all non-COVID diseases for a single week, starting 1/1/2017 through 8/1/2020. The orange line shows the 95% confidence level for each date: if total deaths exceed this line, there’s less than a 5% chance that this occurred by chance. A red + on the chart means this confidence level was exceeded. The blue bars show reported COVID deaths separately, which get added to deaths from all other diseases.
Both New Jersey and Florida reported their first COVID deaths within a week of each other. Florida’s deaths stayed at a low level for a couple of months, and didn’t “take off” until after Memorial Day. Even then, the take-off appears relatively modest, a fraction of baseline.
New Jersey’s COVID deaths exploded almost immediately after the first death was recorded (so that combined deaths peaked at nearly 3x baseline).
In both states, after “take off,” you can see weeks where non-COVID deaths by themselves exceeded the 95% interval. Given the location on the timeline, these should be considered “probable COVID deaths” which weren’t included in the original COVID death counts. For Florida, we estimated 3,973 Probable Covid Deaths during 2020 through August 1 (Note 3).
Taken by itself, adding this estimate of Florida’s probable cases increases the CFR from 1.6% to 2.4%. Of course, it doesn’t stand alone.
NJ was Unlucky
Overall, our comparison of Florida’s vs. New Jersey’s CFR, and normalizing for age differences, is as follows:
Source of Adjustment
“Naive” CFR Calculation
Adjustment for Florida age distribution
Adjustment for newer cases
Adding in Probable Deaths
Estimated, normalized CFR
Pandemic-Sense.com original analysis
So far, therefore, we’ve brought the difference between NJ and FL from 5x to 1.8x, which is much less than before but still a big differential. Why do differences remain?
In my opinion, there are a number of factors. As we’ve seen, only some of it relates to the quality of hospital care. That said, most of Florida’s cases are hitting 70 days later than New Jersey’s. That’s an eternity in Pandemic-time, and means — all other things being equal — that survival rates probably are higher right now in Florida compared to New Jersey last April or May. New Jersey hospitals are probably doing equally well or even better, but there aren’t enough new cases to improve the CFR very much.
In addition, when you assess New Jersey, you can’t ignore that it has the highest population density of any state in the US, and is located next to New York City which started as the epicenter of the initial outbreak. Clearly, if you look at the excess death plot, there was initially much earlier community spread in NJ compared to Florida, and a much greater number of un-diagnosed cases.
Particularly bad for its impact on death rate, there was a much more extensive, early spread of the disease into NJ long term care (“LTC”) facilities.
LTC Residents (2019)
Licensed LTC Facilities (2019)
LTC Facilities with Clusters > 50 (7/12/2020)
Confirmed Cases in LTC Clusters (7/12/2020)
Sources: 2019 LTC data from Kaiser Family Foundation analysis of Certification and Survey Provider Enhanced Reports (CASPER) data. Cluster data from NY Times, 7/12/2020 as classified and geo-located by Pandemic-Sense.com. (Note 4)
In the wake of COVID-19, there’s been a good deal of finger pointing in New Jersey between state regulators and the LTC industry, suggesting poor judgements (e.g. hospitals given priority for PPE), to little oversight, and insufficient/incompetent infection management. Be that as it may, given this once-in-a-century event, it’s not unreasonable to suggest that problems were at least partly due to bad luck. How bad NJ’s situation is compared to Florida is emphasized in the map below (both states plotted to the same scale) which shows the NY Times LTC Clusters.
Overall, our hospitals and front line medical staff are doing amazing work to preserve the health of the US population. They deserve better support than they’re getting.
(Note 1) These low fatality rates should NOT suggest that it’s safe for young people to catch COVID. Long term health effects are unknown for all ages, but there is significant evidence that young people may suffer debilitating morbidities at a much higher rate than fatalities. Young people who get infected can also expand the pandemic by infecting others who are more vulnerable.
(Note 2) There is a tiny, very long tail for COVID fatalities. For analysis purposes we’ve ignored it, but anecdotal reports of people dying from COVID after illnesses lasting 90+ days are not uncommon.
(Note 3) pandemic-sense.com original analysis based on CDC Florida Excess Deaths with and without COVID-19. We examined weeks where non-COVID deaths exceeded the 95% confidence interval by themselves, and estimated probable COVID deaths by calculating the difference between non-COVID deaths and the CDC’s expected deaths from all causes.
(Note 4) Pandemic Sense classified the NY Times cluster data based simply on the name of the institution. The Kaiser LTC data is undoubtedly a somewhat different universe and is not directly comparable (which is why we didn’t calculate percentages combining the two datasets).
On July 12, the New York Times published a list of COVID-19 clusters numbering at least 50 cases, and has continued to update them. The Times publishes this as a flat list of locations and towns. We were able to automate the geo-location for nearly all of the addresses on the July 12 list and generate a map. Unfortunately, we’re not in a position to update this regularly, though we may from time to time. Zoom in to see the precise location and click on the place-marker for more information.